Chipflation, Agentic DevOps and the New AI Governance Problem

Hey, everybody. Welcome to the Techstrong gang. It's Friday, and my Fridays are always the same.

" And I'm pretty sure we all experience the same thing every week. But we have a great lineup today, talking about some interesting events this week, and there's been a lot of subtle things happening, but they're profound when you start thinking about them. Let me introduce our panelists today.

Jack Gold, how you doing, buddy? I'm good, Mike. Thanks.

Good to see you. Jeff Rich, how are you? Doing great.

Happy National Peanut Butter Cookie Day. Yes. Well, I'm sure we'll figure out how to work that into this conversation, but it will make sense eventually.

Tracy Ragan, as always, good to see you. How are things? Thank you.

It's been great. Glad to be back after a two-week hiatus. We missed you.

And John Schwartz, good to see you. And how are things, my friend? Well, it's 58 minutes until SpaceX starts trading.

Yeah. Which is going to be crazy and probably be a subject we will tackle Monday- On Monday ... when we all come back.

Yeah. There's just too much fabrication in there to sort out instantly. So we will take the weekend to sort that one out.

But let's get into this whole thing of, I think it's Morgan Stanley, right? Morgan Stanley has a report talking about chip inflation. And what it's pointing out is that the cost of all these processors and memory is going up substantially, and that's having an effect all across the IT ecosystem.

And it seems like it's getting too hard to even plan what's going on here because you don't know what the cost of something's going to be three months from now, six months from now, and I don't see any relief in sight here. But Jack Gold, this is a 66-page report that describes what I just said in about, I don't know, 35 words. But what's going on here from your perspective, and is this going to ever change?

Is this our new reality? Well, never say never, but what is really going on right now in the marketplace is that we're all paying the AI tariff on semiconductors. And I don't mean that from a Trump perspective, right?

He's not putting this on there. But what's happening is that with the incredible amount of demand on chip supply right now, you can only increase demand so much in a short period of time. So as an example, memory.

Memory is a huge issue when you're putting together AI compute systems. And there's only basically three memory providers out there of any size, Micron in the US, SK Hynix, and Samsung in Korea. And they've been providing our memory chips for our smartphones, our PCs, consumer goods, whatever uses a chip, and pretty much everything uses a chip these days.

AI systems use HBM, high bandwidth memory chips, which are extremely, well, today extremely expensive because they're in very short supply. They're much harder to make. And so what these folks have been doing, the memory producers, is switching over from the traditional commodity memory chips that go into your phone or into your PC or whatever else, your TV, whatever it happens to be, to these high bandwidth chips, which are harder to make and are in great demand, so they've increased prices and, of course, that means the margins go up.

So do I produce low-cost, low-margin chips for your phone, or do I produce high-cost, high-margin chips for AI? Same is true of processors. Do I create processors for your phones that sell for $10 or $15, or do I help Nvidia build chips that sell for $30,000 or $35,000?

And TSMC has just so much production capability. The further problem is that in the semiconductor industry, it's not like you can just bring up another factory. Even if you started today, it's two to three years at least before new production capacity comes online.

And so that means that we're not going to see any real help in this space, of course, unless the market crashes, which we all hope it doesn't, the bubble bursts, until new capacity comes on board. And that's a two or three-year issue. So what's ultimately going to happen if you're a company and you're trying to buy 10,000 new PCs this year, they're going to be more expensive.

They're going to be harder to get. It's going to take longer to get them because HP, Dell, Lenovo, they're big companies. They're able to get some limited supply over some of the smaller companies, but it's going to be increasingly harder for them to get them and more expensive.

And if you're a small supplier in the world, you're going to have a real problem trying to get all the memory and CPU chips and GPU chips that you need to produce stuff. So this is not going to be a short-term issue. Again, hopefully it doesn't happen.

If the market crashes, then that's a whole different issue because then you've got supply. But this is the new oil. If you cut off supply, if you shut down the Strait of Hormuz, people aren't getting oil.

The prices are going to go up. Same thing is happening in the chip space. So bottom line, if you're a company, you need to start planning.

Even if you're an individual and you need to buy something new this year, you need to start planning for 10%, 15%, 20% increase in prices Does that mean I buy now and I store it somewhere on the assumption that I'm going to use it later because the cost is going to be higher? Some people are doing that. The problem is that it's hard to get that to even store right now.

You could buy food and stick it in your freezer, right, and eat it in three months. But if you can't get the food now, having a freezer doesn't do you much good. There's a real supply shortage, and that's a big part of the problem.

But the big companies are pre-buying. They did pre-buy, and they're going through their inventories now, and eventually, that inventory wears out, and then what? And so that's the real dilemma that we're seeing.

And correct me if I'm wrong here, but whatever new supply comes on in the next two or three years is probably going to be also aimed at AI requirements. Sure. And so therefore, I may not see any relief on this issue for quite some time because whatever is going to be available is going to be higher-end stuff aimed at AI versus the everyday servers and PCs that 99% of the rest of the world is relying on.

Indeed. If you're a manufacturer, do you build parts that sell for $300 or sell for $3? Yeah.

It's a pretty easy decision to make, and that's what's happening in the marketplace right now. Right. Yeah, I think what we're seeing, though, is the companies who have the money and have guaranteed access to the memory and the compute power, they're going to move faster while everybody else may be priced into kind of a slower AI adoption.

And maybe that's not a bad thing. Maybe that is not a bad thing. If some companies are priced into the slower AI adoption, maybe it'll give us some time to figure it out.

Yeah. But the geopolitical situation could get worse. We already know that any kind of disruption in the supply chain, export restrictions, sanctions, shipping problems, and of course, China and Taiwan tensions, could really cause chipflation, as they're calling it, to become much worse.

And that's going to impact more than just AI acceleration, but it's going to impact all kinds of consumer goods. You were going to say about the consumer side, the memory chip prices are up what, like six-fold in the last year? That's what it said in that article.

That was- Yeah, I mean that- ... shocking ... just was kind of stunning.

And in a sense, you think of what the downstream hardware companies are going to be absorbing this hit, which will be passed on to the consumer, higher prices for smartphones, PCs, cloud services. So I mean, correct me if I'm wrong, but the semiconductor industry is a boom or bust industry, or has gone through those cycles, and now we're talking about this permanent structural shift in essentially the global technology supply chain, right? Well, I would argue it's not permanent, because what ends up happening is that as prices increase, new supply comes online, just because it's more attractive for them to do so.

And in this particular case, I think what ends up happening as a derivative is that companies in China who have been wanting to be major semiconductor producers will have even more incentive to move- Mm ... quickly into this space. Not just China, probably Europeans as well, although that's a whole different issue because they just move so slowly because of regulations and other things.

So the supply will pick up, but again, this stuff doesn't move quickly. It's not like you can just put up a warehouse in two months. Yeah.

No. You can't even buy semiconductor. It's not just the semiconductors, it's the semiconductors equipment that makes the semiconductors that's in such short supply.

And- So does this create-- Oh, can I just ask one quick question, then I'll get out of the way. So does this create, Jack, this permanent class divide in tech procurement between the big three like Samsung, Micron, SK Hynix, and then the traditional hardware companies? What it does is until the market stabilizes, until we can get back to a more rational...

The semiconductor industry has always been cyclical. They go up and down depending on what's happening. We're in a peak now.

The question becomes how soon before we get to a more level operating environment. And with the AI peak, again, if there's a bubble, that's a whole different issue because then people just don't buy stuff. But right now, they're selling everything they can possibly sell and at premium pricing.

Until that changes, until supply comes online, I don't see it changing. I'm with Jack. Everything's permanent until it isn't.

Right. And we had a premonition of this six years ago, if you remember. For a different reason, slightly different parameters, but if you remember, during the COVID peak of the COVID problem and the whole supply chain issue, we felt the same way here for a different reason.

I'm not saying the solution to that's going to be the same as this, because it isn't, because this is truly demand for new chips versus simply supply chain clogging everything up. But it is both of those, and it's more than chips. I think chipflation is a symptom of the larger inflationary costs around AI, because I'm certain we're going to talk about later today the whole cost of using AI and the fact that what we thought was going to be a cheap commodity is actually a very expensive cost, with a cost that keeps rising.

So you combine all these together, I think the whole AI picture's going to start looking different, and it's no longer going to be as easy as ask ChatGPT a question. Really, it's going into what does AI really cost us short-term and long-term? Yeah.

And it's not just- We're going to- It's not just AI, Jack. I agree with you 100%. But the typical cost today, estimates of the typical cost in a new car, about half the cost of a new car is electronics.

Mm-hmm. So if electronics goes up by 15 or 20%, what does that do to cars? What does that do to your refrigerator, which are all Wi-Fi connected and processor based today?

And so this is going to be very widespread. It's not just about the AI issue. Of course it is, but that's the root cause.

That's not the result of how it's going to affect us all. All right. So let me ask this question.

Oh, no. Sorry, Tracy, go ahead. During the Biden administration, we all know about the CHIPS Act, and we had discussions about the CHIPS Act.

And on a Techstrong women interview with Jamie Thomas, I asked her about that CHIPS Act, and she said it was essential. We had to have it, but it was going to be 10 years out before we'd see an increase in chip manufacturing, at minimum, and it was going to take something like $60 billion in investment. So this isn't an easy cycle to fix, right?

It's an expensive cycle to fix. To bring on these new manufacturers and to make the corrections to Intel has taken us quite some time, but I would hope that we are going to see a change in five to six years. Maybe we're halfway through that.

Yeah. One more quick comment on that. I saw a new report this week, and I don't remember the exact numbers, but basically they said that even if we built chip plants, there is going to be a shortage of about 100,000 technicians and engineers to run these that we can't build.

Yes. That's what her point was. You got to train in the United States.

We have to train people. Right. That we don't have that kind of expertise coming.

We'd have to bring people from Korea to do the training, or Taiwan. Mm-hmm. It's a major undertaking.

So let me ask Jeff this because, well, I've just been dying since I've been holding my tongue this whole time. But, Jeff, is there something to do with this about public policy? Because it ultimately sounds like this is an AI tax that's being levied on all of us, or is this just the way the cookie crumbles?

Well done. Well done. Wow.

Oh, ouch. I think I would love that if it were that simple, Mike, but I don't think it is, because as you've just heard, it's more than just the cost of chips, which have a ripple effect across life now. There are very few things you do that don't have a chip associated with an activity.

So first, there's that one. Then there's a cost of the more effective we think we're making this, and I say think on purpose, the more that costs us. Add that to the fact that the more we use AI, the more we're depending on all this, it becomes a cumulative effect, which does raise the cost of doing all of it.

So is it truly a tax? I don't know, because I don't know where that tax is going. Not that I know where every tax from the government goes.

That's a different story, but I'm not certain we can say, "Here's where that tax gets deposited," because it's distributed throughout that entire network, starting with the hardware manufacturers, or the chip manufacturers, or the software companies, or the distributors of it, the aggregators of it, the LLMs. Add it all together, and I think we're seeing that AI inflation I was talking about, and Jack mentioned as well, that where equipment drives costs up. Not to mention the fact that I think we're at the point pretty soon where the effectiveness of AI is going to start diminishing.

Because even though we think AI is getting smarter and smarter, I believe what the LLMs are getting is more and more of our data, which is completely corrupt. If our data were accurate, we wouldn't want to need AI. So I believe AI is probably going to start, at some point, peaking and start getting dumber and more like us.

Now, that may be an outlier position, but and the more that happens, the more it's going to cost to try to make it work right. It's already- Hey, Mike. Hey, Mike, can I mention something?

I think you brought this concept of an AI tax. You don't suppose we're going to start seeing politicians running in the midterms who are going to start oversimplifying this and conflating this issue? And maybe they're going to be right in a certain extent that these are unknown costs or taxes associated with AI and the evils of AI.

Mm-hmm. You can just see this coming, along the data center argument. Oh, hold on.

It'll get better than that. So yes, that will definitely be the case, but this too will happen. About the US consumer has been holding up this little economic engine of ours, and the US consumer- Mm-hmm ...

is not prepared to absorb 15% to 20% price increases across the board on refrigerators, cars, and everything else that they're consuming with electronics. So you can probably see that this is going to be not just an AI issue, but it's going to become an economic and political- And economic, right. 2% or wherever it is, they're going to stop spending, right?

They can't afford this. Right. They got to put a stop, and that's going to impact the economy.

Yeah, you're right. It's this ripple effect. Right.

And the companies who are spending more on this gear are not going to be in a position to give them a raise to compensate for all of that. So at some point, this all comes to a head, probably in about 6 to 12 months would be my guess, but I don't know. Maybe I'm crazy.

Tracy, am I crazy? Not normally. Sometimes you are, but on this one, I think you might be spot on.

It could be sooner than 6 to 12 months, though, depending on our geopolitical situation, and that's a wild card. There's another ripple to this, by the way. I don't know if you guys saw these headlines this last week or two.

They shut down hundreds or thousands of social media Channels that were basically run by the Chinese and the North Koreans and the Russians, whoever else, trying to hit people up about how bad AI will be for us. About bad for the environment, bad for cost. It'll say nasty things about us, to your point, Jeff, about how crazy it could possibly get.

And so, to your point, John, we're already seeing it. That's interesting that you say... This is a crazy comparison on my part, so please take it with a grain of salt, but you mentioned this.

I'm glad you mentioned that, Jack, because in a sense, it's like messaging. And I even think back to the Vietnam War, and this is a stretch, but in terms of messaging and marketing and sending a message to the American public and propaganda- Yeah ... kind of seeing similarities.

Yep. Absolutely. Where's Walter Cronkite when you need him?

Well, it's just another reason to hate- Oh, you're old, Jeff. It's another reason to hate AI, because you have these hyperscalers crowding out all the traditional buyers. Right?

They're locking up long-term supply agreements, and they get priority access. But the consumer electronics and traditional enterprise IT and smaller hardware vendors are the ones who are fighting for this constrained supply. " Well, and there was another story this week, again, quickie, where they found that AI trained on human behavior, which is where it all gets trained, right, is now picking up the same biases that humans have.

So if you're Black, if you're the wrong religion- Of course ... if you're LBGQ, whatever it is, now AI is promoting that stuff. So, where does it stop, guys?

It's a real issue. And politically, it's going to be a huge issue going forward. So I'm going to say it again just because I want to be on the record.

These foundational LLMs are not the answer. Right. I think we still have to get back to small language models, domain expertise, and stop thinking that we're going to have three or four foundational models that everybody does everything on.

And I hope that the US government is listening because they're the ones that need to hear it the most. Good luck, Tracy. I'm going to have to move this conversation along because it could go on forever, but I would just point out that in this political season, you've been hearing the word affordability a lot.

Well, you're going to hear the word affordability and AI in the same sentence a lot as we go forward. And mark my words, it's all coming. It'll be here in the next few months.

All right. I do want to shift a gear here, though, and there has been some amazing things happening in the land of DevOps. And I would argue that maybe we're on the cusp of some sort of renaissance period in DevOps, but I'll give you a couple examples.

GitLab went out and figured out that the AI agents were overwhelming the Git protocol that makes all these CI/CD platforms run, and so they've come up with a new flavor of it that scales better, and it's based on a more distributed architecture. AWS, at the same time, has figured out that, well, all these harnesses that we're using are tied to a specific vendor, and so we've kind of come up with an open source approach to a tool that creates a framework that anybody can borrow to go create their own harness that is not tied to a specific LLM. And then we saw Datadog come out and extend the reach of their observability platform, where they're actually including some coding tools that would actually solve the problems that they discover for you in their observability frameworks.

That's just the tip of the iceberg. There's all kinds of fun stuff happening, and that's just one week's worth. So I don't know.

Tracy, as I look at this stuff, we've been complaining about the volume of code that's been generated, and we're not managing it especially well, but are we on the cusp of something new and different here? I think we are. It hasn't been that long since I got back from Minneapolis from CDCon, and we did three areas of topic.

We did security, platform engineering, and AI. And the platform engineering and the AI sessions, all of the sessions were well-attended, but those two had the most questions around them. So I think that we are starting to see a clear shift in the DevOps market, that the DevOps platform itself is going to become, for lack of a better word, a control plane for AI agents.

And I think these announcements, the six that were pointed out, all point in the same direction, that AI is no longer just helping developers write code inside an IDE, but it's starting to be embedded into systems that observe applications, review code, enforce policies, remediate vulnerabilities, hopefully. We'd like to see more of that. We'd like to use some of that in our platform.

But I think the real question here is, and I ask this often, is that are DevOps teams ready to manage AI agents in the same way that they manage a human contributor? Because everything we have done has been based on that. I don't even know if developers are doing proper code reviews of what comes out of generated code.

They should. We should be treating AI-generated code like a first-time college student coming into a company and starting to write code. We have to deal with identity and permissions and policy and evidence and observability and rollback in these DevOps pipelines.

Because once agents start writing and reviewing and remediating this code, the risk moves from did the AI generate good code, or who is governing the autonomous workflow? There are some questions here that we don't have answers to. Mm-hmm.

And we are slow at solving them because many of our DevOps professionals are happy with their Jenkins work posts, to be quite honest. And I'm hoping some of these new features coming up at GitLab, for example, will encourage more innovation in the DevOps marketplace, which we haven't really seen in quite some time, to be honest. And to put a finer point on it, we are hitting some sort of tipping point here, and it comes in the form of the first generation of the AI coding tools were trained on flawed code pulled from all over the Internet.

So the quality of the code that they're generating is suspect, and how much of that is getting into a production environment is unknown. And now we have these more advanced AI models, such as Mythos, that's making it easier than ever to discover those vulnerabilities and also come up with the exploit. In fact, the exploit's coming up sooner than the actual patches.

Mm-hmm. So, at some point, it feels to me like we've reached the point where we have to go and fix our DevSecOps workflows, and that means new tools, new platforms, new training, and we're basically going to have to start this whole thing over again. But is that- 100%.

The DevOps platform has to become the system of record, and the system of control for AI-generated change. At CDCon, I even brought up the prompt injection issue, and Brett Smith from SAS indicated that they were trying to write versioning for prompts. So something as simple as that we haven't achieved.

Right? So we have a long road ahead of us, and I do think that it's going to require a complete revamping of the DevOps platform, and that might come from the platform engineering side. But I think what we're seeing is SRE, the role of the SRE is becoming automated through AI, and that may not be the best place for it because we're taking too much of the human out of the loop, and we have thrown away some really solid steps that have kept just human coding clean.

That we're just making the assumption that AI generates good code. I don't know why we do that, but that's what we're doing. And that changes the game pretty big.

So the Datadog's observability is going to be pretty important. And every agentic AI process needs governance, not just productivity metrics. I know that's important, but we have to build governance around it and audit records, because if we don't recognize that agents are having a problem, they just become a passive tool, and we don't ever look at it.

And that's not the way to go. Yeah. This takes me back a lot of years, but I remember when the first Office programs were coming out, with word processing and spreadsheets and all of that, and everyone said, "This is great.

" Guess what, guys? I can give you the tools, but if you can't write, if you don't have any good ideas, Word isn't going to help you make a book that people are going to read. And we're getting to the same point now with the DevOps stuff, right?

At the end of the day, these are tools, and if you don't have people who know how to do it the right way, they're not going to help you. It's going to make things even worse. Jeff, a lot of us have a bad feeling in the pit of our stomach, and I'm trying to get your read on this, but it feels like we are now generating more code that is flawed than ever, and the bad guys are getting better at exploiting it faster than ever, and maybe in ways that are more sophisticated than we think.

So is this just a recipe for a disaster? It potentially could be, but I don't think it's all gloom and doom. We need to keep in mind, on our Fridays, Joe, we need to add some brightness to it as well.

Right. It's not all terrible. But in that same vein, you're right, Mike, your observation's spot on.

Now that we have these tools, we're creating zero days much faster than we used to, which is why you're discovering the exploit before you discover the patch. And these same tools are in the hands of our adversaries who can find them faster than they used to. So even though the overall picture doesn't change, the scale and speed does now with that from a security perspective.

And Tracy talked briefly about the identity of it, and Mike, you referred to it as well. " Because, one, it's not a person, thankfully, and two, even as smart and as effective as these new AI tools are, and by the way, these are all good news. I like the observability.

All of these are great steps forward. We don't need to fix our DevOps program. We need to destroy it and start from scratch because the world is different.

We need to integrate AI into what we do. I think it's a very good thing. We need to make sure we don't exclude humans from that loop because we are necessary for what we want to produce.

We haven't found that blend yet. And Mike, to your question, no. I don't think many, if any, DevOps organizations are well-tooled and educated and disciplined enough to run AI-generated code on its own and manage it.

We are not there yet. It's not just the code, right? It's the LLMs themselves.

Yes. It's a overall change. What is DevOps?

It's change management, right? That's in essence what DevOps is. It's how do you manage change coming across the enterprise?

And we always think of managing change because of source code and a database update. But now we have source code that's being generated quickly, and we have models that we don't have any control over. We can't manage the change in a model.

We can't see if we've got memory poisoning. We can't control that at all. So we are injecting a very critical component into our software that we're delivering that we can't manage the change around.

That is, in essence, the problem. Just to break it down, we cannot manage that change. We no longer have change control practices around AI-based software.

It's incredibly hard to do, especially if you're using a foundational model and you're not building your own small LLM to build that AI intelligence. You can't manage it. It's really impossible right now.

So I missed that ray of sunshine that Jeff was talking about. I know. I'm sorry.

I just have to say, I break it down to being a change management problem. That's what we're facing. Mm-hmm.

It's as simple as that. But here is my hope. My hope is that we will get through this, and that at the end of it, the software we do deploy is going to be much higher quality, much more secure than what we have been shipping all these years.

And we've been lucky, I would say, that we haven't had as many calamities as we probably deserve. But is that a possibility in a two-year window, maybe three-year window, we could wind up with better software than ever? What do you think?

I don't know. Jack, I haven't heard from you. What do you think?

Look, at the end of the day, I still think it's all about supervision. Supervision in the training of the models that will build the code, supervision in the sense that people need to oversee what's going on, supervision in the sense that before I deploy, I need to test a lot more than many companies do today. And so yes, Mike, over time, things improve.

They just do, unless we're really stupid about things, but they do. Do we get to some point where AI generates better code than a human does? Perhaps.

But I don't think we're anywhere near that right now, as Tracy said, as Jeff said. There's a lot of stuff that needs to go on between now and then around making the models better, making the process better, understanding where the models are being trained. Are the models being trained on bad code, which then produces bad code?

Or are they being trained on good code, which then hopefully produces better code? So there's a big feedback loop here that I'm not sure we're totally in sync with yet, and I think that's going to be the real key to making this all work. Can I offer an injection here of where...

I like living in analogies, and if you want to understand a society, sometimes the best place to start is at their landfill. And I'm serious about this. The landfill from where you live says almost everything about how you live.

And I believe, and Tracy, you sparked this thought with me, that the LLMs are our landfills. Everything is getting dumped in there. Exactly.

That's the point. It's foundational, so it's just everything. There's no domain expertise.

Mm-hmm. So we do need specialized language models that are local to what we want to do, rather than always going in and taking a shovel full of whatever's in the landfill and assuming that's going to be accurate. I love that, yeah.

So Tracy, do we also need to change the emphasis? Because all the emphasis on DevOps, as long as I can remember, has been on speed, and it's all about we want to build the next feature, regardless of the fact that 80% of the features in software probably aren't consumed by anybody, but we're going to add more features faster than ever. And maybe the time has come to think more about the quality rather than the actual new feature that we're racing out the door.

And furthermore, is it also to the point now where maybe we should be thinking about safety and just raising the standard for the type of software that we want to deploy and what's acceptable? All of those things, Mike. All of those things.

The problem is that I don't see companies focused on investment in DevOps in the way that they did when DevOps was a buzzword. And right now, we need a new kind of DevOps for these new kind of applications. Our old processes, our old workflows are going to still be out there.

They're going to do fine. The problem is we're not creating fast enough new tooling to support what we need to do for change management through the DevOps process. We don't have those tools even to add to the workflows, right?

And our workflows have to learn themselves. So I don't think that there's an investment going into solving this problem from large companies that usually guide us through this process. You get the big companies that are starting to do really cool DevOps stuff.

They talk about it at conferences. Everybody learns from that. I didn't see any of that at CDCon this year.

I didn't see any of it. So I think that the teary side is there's a great opportunity for building new DevOps platforms and tooling. And I think that the announcement that we're seeing is a beginning of that new tooling, but the tooling's useless if we don't build it into new DevOps workflows.

And companies just aren't spending money in building out new DevOps platforms. They think that they've got it solved with the current tooling, and they don't. So Tracy, I'm with you, but one of the problems that companies have always had, no matter what the technology is, they don't like investing in overhead, right?

Oh, I know. That's the fundamental issue. Even in the age of AI, it will become more expensive to deliver quality software.

This is true. But hey, you get the software you deserve at the end of the day. And let me just say, they don't want to invest in change management.

They want to invest in getting product out to customers faster. Right. Right?

And now we're addressing, we're back to dealing with change management issues. There you go. And getting the product out to faster means generate the code using AI and push it forward.

" And I'm like, "No! " Right. No vibe coding.

I got to shift this gear because we're running out of time for our third episode here, but there's a new report out from IBM talking about the fact that, well, the IT folks don't feel like they have control over AI. This may come as a shock and a surprise, but the issue here is, and I'm going to toss this to Jeff, is it seems to me we started out early on trying to make sure that IT people were not involved in AI because they might get in the way. And now we've kind of reached this point now where we're worried that the IT folks are not involved in AI because, well, things are getting more costly and spinning out of control.

So is this an inevitable outcome, or what's to be done here, Jeff? No, I believe the report is spot on. I think it's missing one key factor, though.

IT has not been in control of IT for at least 15 years. It's an illusion? It is.

And it started with, it used to be, and Jack, at the very least, you and I, I'm sure remember when everything happened within the data center and IT controlled every single bit of data that came in and out. Right. That stopped over 20 years ago.

Yeah. 15 years ago, when the cloud really started to become, starting with Salesforce and others, becoming more prolific, IT really thought they were controlling everything, and they figured whatever became shadow IT, like someone in marketing getting a Salesforce account, was no big deal because they still controlled what happened in the data center. That's when that illusion of control started, and it's been maintained ever since.

Now, I'm not certain that's all bad news because, in my opinion, IT cannot run a business. Every time I've seen it try, it usually fails. The business has to run the business, and IT is certainly a part of it.

IT is a huge business component, a large tool. I don't think business in general has recognized that yet. This IBM report really talks, I think highlights that, saying AI is now running and there's bad stuff happening that when it occurs, they're either blind to it or they tell IT to fix it.

" You're going to pay a lot for it and you're going to have a longer outage, and you didn't do yourself any good. That's all time you'll never get back, and money and damage to the house and everything else that occurs. As I said before, I live in analogies.

That's kind of where we are right now, though. And the problem is the damage to the house are disclosures, are new exploits that are being exposure-- vulnerabilities are being developed and information that's being leaked. " And I think AI is driving a lot of that volume and speed.

So IT no longer controls it. Just like I said DevOps needs to blow it up and start over, I think IT needs to do the same thing. Technology is no longer managing the data that goes in and out of the data center.

Technology now has to be what tooling, processes, and governance do we need to have in place to make sure all of our data flows are working the way we expect them to. It's a very, very different picture. " The AI agent doesn't have intuition, it doesn't have experience, it hasn't seen other similar problems before.

So we need to manage AI agents differently. There are companies that are selling tools now that can help you with that, but we're still not there yet. Are you- So Mike- Wait, wait.

Let me just- I'm going to interject something because... Oh, jeez. As these IT studies go, usually my eyes glaze over.

But this one has some pretty compelling numbers that I think I just wanted to throw out there just to whoever's watching, because these numbers are pretty stark. What is it? The IT budget that's being gobbled up by AI costs is going to go from 15% last year to 25% in 2027.

And yet despite that allocation, almost 85% of the tech leaders have very real lack real-time visibility into these costs. It almost screams to this idea that these companies, and IBM advocates this, they need to build automated control and visibility into their AI infrastructure. They don't have a choice.

Mm-hmm. Well, I'm going to go back to Jeff's metaphor. So Jeff, is this landfill that you were describing, is it actually sitting on top of a large amount of cesspools where all this data kind of that you described, where all this data kind of is just like swirling around in some muck somewhere, and then we built this LLM on top of this stuff, and away we go?

Well, I'm so glad I brought that analogy up. If we want to follow that, the answer has to be yes, because landfills, at least parts of them, do compost. back into the land, and then you have true garbage left over at the end.

And if we want to take this analogy to its extreme, I think that's the course we may be on. LLMs were a great, big, powerful resource to fuel, to use to get AI going. I don't think we can count on it anymore.

Well, let me rephrase that. I'm not looking for a specific end to stop using them, but I don't think we can count on them to help us grow further because we can't get enough specialized data out of there. Are we picking up garbage?

Are we picking up something that's compost, or are we picking up something... By the way, landfills sometimes have valuable things contained in them as well. So do we know what we're getting?

Until we get specialized, smaller language models that we have more control over using, we can't count on that. So yeah, maybe it is assessable. Maybe there's some sort of new monster that's being grown because some alien bacteria came into the landfill that we don't know about.

So Mike, I'm going to stay with where you're going on this. It is a problem that we cannot continue with. We have to find a way to get more specialized tooling.

I'm sounding like Tracy here. Tooling, processes, and data that we know we can count on so that AI really can help us into the future. There is another piece to this, guys, and it's foundational.

One of the reasons for shadow whatever, right? Shadow IT, shadow AI, shadow whatever, is that many businesses see IT as an obstacle. They don't see them as helping them do better business.

They see them as putting roadblocks in the way that makes it harder for me to get my job done. We saw that over the years. Jeff, you're right.

We're old guys. I remember mobile. How did mobile get into companies?

It wasn't through IT. It was because people brought them in because it worked, whether it was BlackBerrys or whether it was a PC, or a lot of other stuff. And so the same is happening with AI now.

Companies are now seeing a lot more shadow AI come on board, so it's harder for IT to manage that. But part of that is because IT is not keeping up with the business need. That's not an easy problem to solve.

That's a really hard problem to solve, and it's not just about governance, it's also about becoming part of the solution to the problem. And we've been talking about that for many years. It's not probably going to change because IT tends to move slowly and businesses tend to want to accelerate and move very quickly.

That has to be part of the equation. I don't know how we fix that problem. All right.

I'm going to leave this here because these are big problems and they will take a long time to fix, but I think we've got to start working on it now if we're going to help to maybe solve any of these issues before the end of the decade. In the meantime, I will just call everybody out. The DevOps conversation was great.

There is a webinar on Monday that I'm hosting at 1:00. Please attend if you are interested in that DevOps topic because we'll be talking more about AI and CI/CD platforms. Also, there is a virtual event in the fall that we're hosting as well, so please take a look for that as well, and it will be about AI native DevOps, and that too will be a great event.

So we'll be talking about this subject for many, many more months to come. Hey, I want to thank everybody for sharing their knowledge and insights as always. How's everybody doing?

Have a great weekend. John, Jeff, Jack, Tracy, as always, brilliant. Guys.

And I want to thank you all for watching this episode, and if you watched anything else this week, thank you for tuning in as well, and stay tuned for the rest of the replay of the Techstrong TV lineup because it too is equally awesome. And have a great weekend, and we'll see you Monday.